Systems thinking: AI in your processes

Beyond the tool: rethinking your processes

Using ChatGPT to do what you already do, but faster, is good. Rethinking your processes by integrating AI as a native component is transformative. The difference doesn’t lie in deployment sophistication, but in whether you’re willing to ask one critical question.

The trap of like-for-like automation

Many AI deployments make the same mistake: automating existing processes without questioning them. It’s like scanning paper forms to email them—you’ve digitized, not transformed. The process keeps its steps, stakeholders, and timelines; only the speed increases, which quickly caps the gains.

Consider a typical example. An analyst receives a 50-page PDF report, spends 4 hours extracting key figures, produces a 2-page summary, and sends it to their manager. Naive automation involves importing the PDF into ChatGPT and asking for a summary: you save time, but the process remains unchanged—the analyst still triggers, writes, and transmits.

True transformation is different. The report lands in a connected SharePoint space; a “Report Analysis” assistant—a shared project or a skill—processes it and produces a structured summary with indicators, anomalies, and recommendations directly in the Finance project. The analyst no longer writes the summary; they validate and enrich the analysis. The gain isn’t measured in hours saved, but in shifting the role from low-value extraction to interpretation and decision-making.

Map, classify, redesign

The method consists of three steps. The first is descriptive: document the steps, stakeholders, inputs and outputs, actual time spent, and friction points—waiting times, back-and-forth exchanges, and re-entry of data. It’s tedious, therefore decisive: most teams discover steps that no one can justify anymore.

The second step classifies each link in the chain into three categories. AI-automatable tasks are those involving summarization, writing, translation, extraction, and classification—such as reviewing a contract to extract due dates. Augmentable tasks involve decision-making, analysis, or creation, where AI assists without replacing: arbitrating between two budget scenarios, for example. Irreducibly human tasks remain strategic judgment, client relations, negotiation, and final validation; naming them is more reassuring than vague talk about AI not replacing anyone.

The third step is the only one that transforms, and it’s the one people skip. Don’t just insert ChatGPT into existing steps: ask yourself if you were designing this process today, with AI as an available component, what would it look like? The answer makes steps disappear, merges others, and reveals new ones—typically a human validation step after AI generation, which was unnecessary when production was human.

What this looks like, function by function

In Sales, reps used to prepare every proposal by hand: client research, writing, formatting. A “Proposal Preparation” assistant, fed by a brief dropped into a shared project, generates a first draft that the rep refines. The effort shifts from layout to argumentation.

In Support, a ticket required reading, searching the knowledge base, and drafting a response. A “Triage and Response” assistant categorizes and proposes an answer based on that base; the agent validates or adjusts, freeing themselves up for cases that merit their expertise.

In HR, recruiters screened resumes, wrote job descriptions, and prepared interviews. A “Recruitment” assistant structures descriptions from the manager’s brief, generates interview scorecards, and synthesizes session notes. The recruiter refocuses on evaluation and candidate relations, while the decision remains human.

In Finance, controllers collected data, built reports, and wrote analyses. Data resides in a shared project, and a “Financial Reporting” assistant produces standard reports while flagging anomalies. The controller analyzes them and proposes corrective actions: the same shift as for the analyst.

Five questions to surface the right processes

Ask the same questions in every department: What are your five most time-consuming processes? Which ones involve writing, summarizing, or document analysis? Which back-and-forth exchanges between people happen most often? If you had an unlimited assistant available 24/7, what would you ask it to do first? And what tasks do you do “because we’ve always done it that way”? The first four identify sources of gain; the fifth, the most uncomfortable, reveals processes to eliminate rather than optimize.

The real obstacle is mindset

Adopting AI in processes isn’t a technical project; it’s a transformation project. Employees shift from an executor mindset—“I do the reporting”—to an augmented supervisor mindset—“I validate and enrich the AI-generated report.” This shift requires trust in the tool, continuous training, and management that values analysis over raw output: as long as annual reviews measure volume produced, no one will change their mindset. Successful organizations aren’t those that deploy the fastest, but those that rethink rather than automate.

Your turn

Redesign a process. Choose a time-consuming process from your team, map its steps, classify each as automatable, augmentable, or human, then draw its “designed today” version. Note the steps that disappear, those that merge, and the human validation you add.

Official sources: Prompting (ChatGPT documentation) · Projects in ChatGPT · Skills in ChatGPT · SharePoint app and setup in ChatGPT (OpenAI Help Center)

Key takeaways

• Automating a process like-for-like quickly hits a ceiling: you must rethink it with AI as a component.
• The method: map, classify each step (automatable, augmentable, human), redesign.
• The key question: what would this process look like if you designed it today?
• The gain is seen in shifting roles, from extraction to interpretation and decision-making.
• The real obstacle is mindset: value analysis over volume produced.